In one of our previous blogs we presented the progress made by pilots toward creating actionable what-if scenarios, otherwise known as epics, to support decision making across three main policy fields: transportation, environment, health. In this article, we would like to elaborate on the traffic, air quality and noise models that will form part of the DUET Digital Twin solution and that ultimately will be used to support the implementation of user scenarios (see D2.3 for a full list) in the participating pilot locations.
Traffic models
Traditionally, specific instances of a wide range of potential transportation models were developed – and in many cases fine-tuned over years or even decades – by specific stakeholders in the domain. A national transportation authority would develop and maintain region- or even nation-wide explanatory models for average traffic loads during peak periods in all transportation modes and infrastructures within their jurisdictions. Such models are often used in various scenarios and cost-benefit analyses related to large infrastructure projects or tax simulations. Local traffic controllers, on the other hand, would develop very precise, minute-to-minute dynamic models of traffic operations on a signalised corridor that they are managing. For them, traffic models offer a way to explore what-if scenarios for better incident response or real-time optimisation for different traffic types.
In theory, the more detailed model layers can be configured for any region or neighborhood in the larger territory. However, the usefulness of such models depends entirely on their empirical validity. At present, no automated calibration procedures exist, and there are generally no commonly agreed guidelines on which data is required to guarantee a certain level of validity of the model outputs. Within DUET, the more refined models will be configurable in principle over the entire territory, but in practice can be trusted empirically only for those zones for which substantial calibration efforts have been performed in the context of specific pilot use cases and using dedicated data sources.
DUET will showcase three different kinds of traffic models: static, dynamic and a local mobility model, which can be thought of as a multi-modal traffic state estimation. Models based on Static Traffic Assignment (STA) capture two basic assumptions that we usually associate with traffic. First, people strive to minimize their travel times by choosing the shortest route, this behaviour eventually leads to something we call ‘equilibrium’ - a state in which all the used routes from A to B have the same travel time, it’s equivalent to saying that no traveller can unilaterally reduce their travel time by switching to a different route. Second, the travel time on a road increases with the amount of people that use it.
With properly calibrated demand data, STA can provide a useful overview of traffic flows in a system and facilitate predictions based on alternate road network graphs or different demand patterns. This covers a wide range of scenarios that could be explored by a Digital Twin. A non-exhaustive list includes
Mobility effects of new developments e.g. building a new apartment complex can alter the mobility demand and cause new traffic from this location to typical destinations (work zones, universities)
Toll-induced deviations from routes on which they are levied
Simulating the impact of road blocks / closed lanes due to construction work
Analysing the effects of changes to a signaling plan
Predicting the changes in travel time caused by larger travel demand to a city because of an event (convention, festival, championship)
Dynamic Traffic Assignment (DTA) brings much more nuance by adding a temporal dimension to Traffic Assignment. It enables us to capture more aspects of the traffic system’s behavior that we observe in reality. Queues are modelled explicitly and spillback effects are taken into account. Traffic Management applications such as Ramp Metering, Route Guidance or signal coordination optimisation need algorithms of their own to find the optimal solutions and may only utilise parts of DTA as a way to evaluate a solution. Those optimisation mechanisms are beyond the scope of DUET, but can play a role in future, more detailed models of a Digital Twin.
Figure 1. Traffic volume (vehicle/hour) on each city link
Lastly, a local traffic flow model (Cityflows) uses different real-time data sources in the city to better understand multi-modal dynamics of city traffic. Based on different data sources (e.g. signalling data, wifi scanning data, camera object detection, Telraam data), the model estimates the density of traffic and discriminates between different modalities, such as motorised and non-motorised vehicles, in the streets of a certain area. By combining Cityflows output data with other available data sources such as OpenStreetMaps and weather API, we will be able to investigate correlations between flows in the city and the commercial areas some of which may also be found in suburbs.
Figure 2. Sample output visualisation of Cityflows
Air quality emission models
Air quality models use mathematical and numerical techniques to simulate the physical and chemical processes that affect air pollutants as they disperse and react in the atmosphere. Three components that are usually quantified in order to indicate local air quality are particulate matter (PM10), ultrafine particulate matter (PM2.5) and nitrogen dioxide (NO2).
In DUET, we’re leaning towards SRM-1 and SRM-2 models given the role that traffic emissions and built environment play in urban air quality. SRM-1 is intended for calculating concentrations of air pollutants near traffic roads in urban areas, also marked as "city roads." Air currents around the buildings influence the air flow in the streets and thus the height of the concentrations of air pollution. This is different to highways and other rural roads where air pollution emitted by traffic does not get stuck between the physical obstacles, but is rather directly carried away by the wind. Standard calculation method (SRM-2) applies to this type of road.
Within DUET’s Digital Twin solution, air quality model emissions will be calculated using data on traffic volume and road network. Other variables such as wind speed and wind direction are also being considered. Eventually, the model will be able to calculate dispersion of air pollution caused by traffic for a grid of spatially referenced calculation points. The results will be converted to map images using interpolation or heatmap technology, and placed on top of a map thereafter. Calculations will be performed for several pollutants (PM10, PM2.5, NO2) based on weather information (wind direction, wind speed) and spatial conditions.
The results of air quality modeling and simulation will help urban developers and city planners assess the impact of policy decisions prior to committing to costly infrastructure changes.
Figure 3. Difference plot for NO2 that shows the effect of a road closure in a city centre
Noise emission models
Traffic noise in urban areas affects the lives and health of many people. To better manage and minimise its negative impacts, the European Commission requires major EU cities to produce noise maps and the corresponding noise-exposure distributions.
In DUET, our intention is to make use of existing open source tools such as NoiseModelling. We already tested NoiseModelling in a desktop environment and are now looking for ways to integrate it into the Traffic Modeller (TraMod) solution. With TraMod, users are able to explore different traffic scenarios by changing multiple road parameters e.g. free flow, speed, capacity. The addition of noise calculation in near real-time would allow them to also explore how, for example, planned roadworks will influence existing noise levels surrounding the construction site.
Another tool under consideration is the Urban Strategy Noise Module. It takes into account three different data sources - road traffic, rail traffic, industry - to create noise maps and various outputs necessary for calculating noise-exposure distributions. In countries like Netherlands it is used as a decision support tool for measuring the effects of different planning scenarios involving infrastructure, buildings and sound barriers.
Figure 4. Noise Lden map for Industry
Next steps: model integration in a Digital Twin environment
This is an innovation challenge that needs to be gradually developed and expanded. Not only does it require a library of modules that in principle should be compatible (e.g. finer-resolution models should be disaggregates of lower-resolution versions), it also requires different data sources and calibration procedures to be integrated in the digital twin environment.
The way in which models interact with one another depends on how tightly they are intertwined. For instance, travel demand depends on the generalized cost and the generalized cost depends on the travel demand. One of the challenges for DUET will be to find a compromise between decomposition of models into their component models (which allows for more intricate interaction schemes and increases developers flexibility) and the effort involved in providing APIs. A decision needs to be made as to what components of larger models should be exposed to the developers using the platform.
It may be necessary to provide the expert user with an opportunity to design their own model-use ‘cookbooks’ where they indicate calculation sequences of different models for some analysis. This is similar to the procedure sequence PTV uses in their macroscopic modelling Software Visum. Alternatively, we may provide external access to the different modelling tools through an API.
From a user perspective it’s reasonable to expect that traffic flows predicted by the assignment models somewhat mirror the different observations made by the sensors visualized within DUET. An interesting extension to DUET would be a (semi-) automatic calibration module that stores and monitors sensor data and compares this with the flows generated by a calibrated traffic assignment module or/and uses statistical methods to identify sensors that are behaving unexpectedly.
If the differences across the network become too large, a rerun of the model calibration module may be needed to update the Origin Destination tables. This should be a carefully considered step though, as calibration has a substantial computational cost.
Comentários